Hi Craig,
On Mon, 2018-06-25 at 09:42 +0800, Craig Ringer wrote:
> The apt.postgresql.org crew have a package morgue for old versions at
> atalia.postgresql.org/morgue/ . It's not a full repo, but you can fish out
> needed packages manually and install them. This is a *lifesaver* when
> trying to examine a core file a customer system where debuginfo wasn't
> installed, or trying to reproduce a subtle version-specific issue.
>
> Do you have anything like that? If not, do you have any interest in it? I'd
> really like to get something like it going, and rather than creating one
> in-house at 2ndQ where nobody else lands up benefiting, it might make sense
> to help out with one for the community.
>
> What I'm thinking of is a selective rsync to an archive repo, where
> packages get copied but never get deleted, then createrepo_c runs in
> incremental mode (--update --retain-old-md-by-age=5d) to index them.
>
> So it doesn't upset https://yum.postgresql.org/, but
> https://yum-archive.postgresql.org/ or whatever can keep a deep history of
> packages.
>
> An alternative is to ditch the repo indexing and use an AWS S3 bucket to
> host an unindexed package slush pile, like the apt morgue. createrepo can't
> run sensibly on s3-hosted files. But s3 is cheap - 2c/gb/month, or less if
> you go for infrequent access mode.
>
> Thoughts?
I'm not against keeping the old package, but there are two things:
* We need to ask sysadmins team, because this means a lot of extra disk space.
* Building older packages is not that hard with the RPMs as you know -- just
change the version number, and run make rpm11 (or whatever). I'm not sure that
keeping the old packages are worth the hassle.
Again, I'm not against the idea as long as you can convince sysadmin team, and
also write scripts to pull the packages :)
Cheers,
--
Devrim Gündüz
EnterpriseDB: https://www.enterprisedb.com
PostgreSQL Consultant, Red Hat Certified Engineer
Twitter: @DevrimGunduz , @DevrimGunduzTR